Image reconstruction of electrostatic tomography based on the improved residual network.

The core of electrostatic tomography (EST) is to solve the inverse problem, but the EST independent measurement data are much smaller than the value that needs to be reconstructed, resulting in a more serious inverse problem. This paper presents an improved ResNet-34 network (P-ResNet), which consists of an input layer, a residual feature extraction layer, and an output layer. The number of residual blocks is 3, 4, 4, and 3. After the second convolution in the main path of each residual block, a ReLU activation function is added to enhance the nonlinear expression ability of the network, and the generalization ability of the model is improved by introducing the L2 regularization loss function. A total of 15 930 sets of samples were simulated for the simulation test. After 200 rounds of iteration, the reconstruction results show that the network achieves high accuracy in EST image reconstruction tasks. In addition, the model is tested under different degrees of Gaussian white noise to verify its anti-noise ability. Compared with the traditional algorithms, the image correlation coefficients of this proposed model network are higher. In addition, this paper designs a small sensor to obtain the induced charge values through the principle of electrostatic induction. The reconstructed results obtained from the experimental data are consistent with the simulation results, which verifies the effectiveness and generalization ability of the proposed model.

Identification of the health education targeted susceptible population of tuberculosis in Ningxia, Northwest China.

Knowledge, Attitude, and Practice (KAP) survey, as an effective measure tool, is of practical significance for identifying the susceptible population in high-incidence regions of tuberculosis (TB). We aim to identify the health education targeted susceptible population of TB and discuss the acting pathway of KAP in Ningxia. A multistage random sampling method was used to conduct a face-to-face questionnaire survey for residents. The latent class analysis (LCA) model was used to classify susceptible populations of TB, and the structural equation modeling (SEM) model was also employed to investigate the interaction mechanisms of KAP (mediation analysis). We further applied the ordered logistic regression model to explore the associated factors. A total of 973 residents were enrolled, 70.6% were male, aged from 16 to 89. The LCA analysis demonstrated that 3 categories of susceptible populations of TB ("overall good", "positive attitude" and "overall poor") have optimal goodness of fit (BIC = 7889.5, Entropy = 0.923). SEM model indicated that the attitude plays a significant mediation effect from knowledge to practice toward TB (an indirect effect of 0.038, and a direct effect of 0.138). The ordered logistic regression results found that age, sex, marital status, education level, occupation, family income, self-perceived health status, having a family member or friend with TB, and knowing the DOTS strategy were significantly associated with classifications of KAP level towards TB. Based on the LCA model, we accurately classified the susceptible population of TB into 3 groups with different degrees of KAP. We found that TB attitude plays a mediating role between knowledge and practice. Therefore, we should pay more attention and carry out targeted health education in the community to these populations with overall poor KAP towards TB, and develop effective strategies and measures to realize the End TB Plan.

Hierarchical Perceptual Noise Injection for Social Media Fingerprint Privacy Protection.

Billions of people share images from their daily lives on social media every day. However, their biometric information (e.g., fingerprints) could be easily stolen from these images. The threat of fingerprint leakage from social media has created a strong desire to anonymize shared images while maintaining image quality, since fingerprints act as a lifelong individual biometric password. To guard the fingerprint leakage, adversarial attack that involves adding imperceptible perturbations to fingerprint images have emerged as a feasible solution. However, existing works of this kind are either weak in black-box transferability or cause the images to have an unnatural appearance. Motivated by the visual perception hierarchy (i.e., high-level perception exploits model-shared semantics that transfer well across models while low-level perception extracts primitive stimuli that result in high visual sensitivity when a suspicious stimulus is provided), we propose FingerSafe, a hierarchical perceptual protective noise injection framework to address the above mentioned problems. For black-box transferability, we inject protective noises into the fingerprint orientation field to perturb the model-shared high-level semantics (i.e., fingerprint ridges). Considering visual naturalness, we suppress the low-level local contrast stimulus by regularizing the response of the Lateral Geniculate Nucleus. Our proposed FingerSafe is the first to provide feasible fingerprint protection in both digital (up to 94.12%) and realistic scenarios (Twitter and Facebook, up to 68.75%). Our code can be found at https://github.com/nlsde-safety-team/FingerSafe.

Surface-Enriched Single-Bi-Atoms Tailoring of Pt Nanorings for Direct Methanol Fuel Cells with Ultralow-Pt-Loading.

Single-atom decorating of Pt emerges as a highly effective strategy to boost catalytic properties, which can trigger the most Pt active sites while blocking the smallest number of Pt atoms. However, the rational design and creation of high-density single-atoms on Pt surface remain as a huge challenge. Herein, a customized synthesis of surface-enriched single-Bi-atoms tailored Pt nanorings (SE-Bi1/Pt NRs) toward methanol oxidation is reported, which is guided by the density functional theory (DFT) calculations suggesting that a relatively higher density of Bi species on Pt surface can ensure a CO-free pathway and accelerate the kinetics of *HCOOH formation. Decorating Pt NRs with dense single-Bi-atoms is achieved by starting from PtBi intermetallic nanoplates (NPs) with intrinsically isolated Bi atoms and subsequent etching and annealing treatments. The SE-Bi1/Pt NRs exhibit a mass activity of 23.77 A mg-1 Pt toward methanol oxidation in alkaline electrolyte, which is 2.2 and 12.8 times higher than those of Pt-Bi NRs and Pt/C, respectively. This excellent activity endows the SE-Bi1/Pt NRs with a high likelihood to be used as a practical anodic electrocatalyst for direct methanol fuel cells （DMFCs） with high power density of 85.3 mW cm-2 and ultralow Pt loading of 0.39 mg cm-2.

Targeted Depletion of Individual Pathogen by Bacteria-Templated Polymer.

Selective and targeted removal of individual species or strains of bacteria from complex communities can be desirable over traditional and broadly acting antibiotics in several conditions. However, strategies that can detect and ablate bacteria with high specificity are emerging in recent years. Herein, a platform is reported that uses bacteria as a template to synthesize polymers containing guanidinium groups for self-selective depletion of specific pathogenic bacteria without disturbing microbial communities. Different from conventional antibiotics, repeated treatment of bacteria with the templated polymers does not evolve drug resistance mutants after 20 days of serial passaging. Especially, high in vivo therapeutic effectiveness of the templated polymers is achieved in E. coli- and P. aeruginosa-induced microbial peritonitis. The templated polymers have shown high selectivity in in vivo antimicrobial activity, which has excellent potential as systemic antimicrobials against bacterial infections.

Association between ambient temperature, PM2.5 and tuberculosis in Northwest China.

Existing evidence suggested that the risk of tuberculosis (TB) infection was associated to the variations in temperature and PM2.5. A total of 9,111 cases of TB were reported in Ningxia Hui Autonomous Region, China from 2013 to 2015 on a daily basis, and 57.2% of them were male. The TB risk was more prominent for a lower temperature in males (RR of 1.724, 95% CI: 1.241, 2.394), the aged over 64 years (RR of 2.241, 95% CI: 1.554, 3.231), and the high mobility occupation subpopulation (RR of 2.758, 95% CI: 1.745, 4.359). High concentration of PM2.5 showed a short-term effect and was only associated with an increased risk in the early stages of exposure for the female, and aged 36-64 years group. There were 15.06% (1370 cases) of cases of TB may be attributable to the temperature, and 2.94% (268 cases) may be attributable to the increase of PM2.5 exposures. Low temperatures may be associated with significantly increase in the risk of TB, and high PM2.5 concentrations have a short-term association on increasing the risk of TB. Strengthening the monitoring and regular prevention and control of high risk groups will provide scientific guidance to reduce the incidence of TB.

Capsule Shedding and Membrane Binding Enhanced Photodynamic Killing of Gram-Negative Bacteria by a Unimolecular Conjugated Polyelectrolyte.

The development of new antimicrobial agents to treat infections caused by Gram-negative bacteria is of paramount importance due to increased antibiotic resistance worldwide. Herein, we show that a water-soluble porphyrin-cored hyperbranched conjugated polyelectrolyte (PorHP) exhibits high photodynamic bactericidal activity against the Gram-negative bacteria tested, including a multidrug-resistant (MDR) pathogen, while demonstrating low cytotoxicity toward mammalian cells. Comprehensive analyses reveal that the antimicrobial activity of PorHP proceeds via a multimodal mechanism by effective bacterial capsule shedding, strong bacterial outer membrane binding, and singlet oxygen generation. Through this multimodal antimicrobial mechanism, PorHP displays significant performance for Gram-negative bacteria with >99.9% photodynamic killing efficacy. Overall, PorHP shows great potential as an antimicrobial agent in fighting the growing threat of Gram-negative bacteria.

[Spatial Distribution of Nitrogen and Phosphorus Nutrients in the Main Stream and Typical Tributaries of Tuojiang River and Fujiang River].

The Tuojiang River and Fujiang River, two important tributaries of the upper reaches of the Yangtze River, have serious water pollution problems, among which nitrogen (N) and phosphorus (P) are the most important pollutants. Therefore, the aim of this study was to identify the influencing factors of water quality in different spaces and provide a scientific basis for the prevention and control of surface water pollution in the upper reaches of the Yangtze River and its tributaries. Water samples of trunk and tributaries in the Tuojiang River and Fujiang River were collected, and the spatial distribution characteristics of water N and P were analyzed. The results showed that the Tuojiang River and Fujiang River showed serious pollution of total nitrogen (TN), with a water quality worse Ⅴ-section proportion as high as 94% and 50%, respectively. Both rivers showed that TN and TP concentrations in the tributaries were higher than those in the main stream. For both rivers, total phosphorus (TP), with moderate pollution, was mainly concentrated in Ⅱ, Ⅲ, and Ⅳ class water quality, whereas the P pollution was more serious for the Fujiang River compared to that of the Fujiang River. For the Tuojiang River, nitrate nitrogen (NN) concentration from upstream to downstream showed a trend of decreasing after the first increase, with the maximum concentration of ammonium nitrogen (AN) exhibiting at the upstream site. In particular, TP concentration increased significantly after rivers flowed through a city. For the Fujiang River trunk stream, TN and NN concentration exhibited a gradually increasing trend from the middle to lower reaches. Generally, our study revealed that TN, TP, and NN in the rivers were affected by water pH and water temperature (T). Therefore, the control of N and P pollution in rivers should pay attention to the influence of water environmental factors.

Associations Between Gender Gaps in Life Expectancy, Air Pollution, and Urbanization: A Global Assessment With Bayesian Spatiotemporal Modeling.

Objectives: It's evident that women have a longer life expectancy than men. This study investigates the spatiotemporal trends of gender gaps in life expectancy (GGLE). It demonstrates the spatiotemporal difference of the influence factors of population-weighted air pollution (pwPM2.5) and urbanization on GGLE. Methods: Panel data on GGLE and influencing factors from 134 countries from 1960 to 2018 are collected. The Bayesian spatiotemporal model is performed. Results: The results show an obvious spatial heterogeneity worldwide with a continuously increasing trend of GGLE. Bayesian spatiotemporal regression reveals a significant positive relationship between pwPM2.5, urbanization, and GGLE with the spatial random effects. Further, the regression coefficients present obvious geographic disparities across space worldwide. Conclusion: In sum, social-economic development and air quality improvement should be considered comprehensively in global policy to make a fair chance for both genders to maximize their health gains.

Diverse Sample Generation: Pushing the Limit of Generative Data-Free Quantization.

Generative data-free quantization emerges as a practical compression approach that quantizes deep neural networks to low bit-width without accessing the real data. This approach generates data utilizing batch normalization (BN) statistics of the full-precision networks to quantize the networks. However, it always faces the serious challenges of accuracy degradation in practice. We first give a theoretical analysis that the diversity of synthetic samples is crucial for the data-free quantization, while in existing approaches, the synthetic data completely constrained by BN statistics experimentally exhibit severe homogenization at distribution and sample levels. This paper presents a generic Diverse Sample Generation (DSG) scheme for the generative data-free quantization, to mitigate detrimental homogenization. We first slack the statistics alignment for features in the BN layer to relax the distribution constraint. Then, we strengthen the loss impact of the specific BN layers for different samples and inhibit the correlation among samples in the generation process, to diversify samples from the statistical and spatial perspectives, respectively. Comprehensive experiments show that for large-scale image classification tasks, our DSG can consistently quantization performance on different neural architectures, especially under ultra-low bit-width. And data diversification caused by our DSG brings a general gain to various quantization-aware training and post-training quantization approaches, demonstrating its generality and effectiveness.

Application of diffusion kurtosis imaging in neonatal brain development.

Background: Deviations from the regular pattern of growth and development could lead to early childhood diseases, suggesting the importance of evaluating early brain development. Through this study, we aimed to explore the changing patterns of white matter and gray matter during neonatal brain development using diffusion kurtosis imaging (DKI). Materials and methods: In total, 42 full-term neonates (within 28 days of birth) underwent conventional brain magnetic resonance imaging (MRI) and DKI. The DKI metrics (including kurtosis parameters and diffusion parameters) of white matter and deep gray matter were measured. DKI metrics from the different regions of interest (ROIs) were evaluated using the Kruskal-Wallis test and Bonferroni method. Spearman rank correlation analysis of the DKI metrics was conducted, and the age at the time of brain MRI acquisition was calculated. The subjects were divided into three groups according to their age at the time of brain MRI acquisition: the first group, neonates aged ≤7 days; the second group, neonates aged 8-14 days; and the third group, neonates aged 15-28 days. The rate of change in DKI metrics relative to the first group was computed. Results: The mean kurtosis (MK), axial kurtosis (Ka), radial kurtosis (Kr), and fractional anisotropy (FA) values showed positive correlations, whereas mean diffusion (MD), axial diffusion (Da), and radial diffusion (Dr) values showed negative correlations with the age at the time of brain MRI acquisition. The absolute correlation coefficients between MK values of almost all ROIs (except genu of the corpus callosum and frontal white matter) and the age at the time of brain MRI acquisition were greater than other metrics. The kurtosis parameters and FA values of central white matter were significantly higher than that of peripheral white matter, whereas the MD and Dr values were significantly lower than that of peripheral white matter. The MK value of the posterior limb of the internal capsule was the highest among the white matter areas. The FA value of the splenium of the corpus callosum was significantly higher than that of the other white matter areas. The kurtosis parameters and FA values of globus pallidus and thalamus were significantly higher than those of the caudate nucleus and putamen, whereas the Da and Dr values of globus pallidus and thalamus were significantly lower than those of the caudate nucleus and putamen. The relative change rates of kurtosis parameters and FA values of all ROIs were greater than those of MD, Da, and Dr values. The amplitude of MK values of almost all ROIs (except for the genu of the corpus callosum and central white matter of the centrum semiovale level) was greater than that of other metrics. The relative change rates of the Kr values of most ROIs were greater than those of the Ka value, and the relative change rates of the Dr values of most ROIs were greater than those of the Da value. Conclusion: DKI parameters showed potential advantages in detecting the changes in brain microstructure during neonatal brain development.

Imaging system design based on electromagnetic tomography for high conductivity medium reconstruction.

Electromagnetic tomography (EMT) is used to create tomographic images of the electrical properties of conducting material based on electromagnetic measurements from coils evenly distributed around the imaging region. EMT is widely used in industrial and biomedical fields for which it offers the advantages of being non-contact, fast, and non-radiative. Most EMT measurement systems are implemented with commercial instruments, such as impedance analyzers and lock-in amplifiers, which are bulky and inconvenient for portable detection devices. In order to improve the portability and extensibility, a purpose-built flexible and modularized EMT system is presented in this paper. The hardware system consists of six parts: the sensor array, signal conditioning module, lower computer module, data acquisition module, excitation signal module, and the upper computer. The complexity of the EMT system is reduced by a modularized design. The sensitivity matrix is calculated by the perturbation method. The split Bregman algorithm is applied to solve the L1 norm regularization problem. The effectiveness and advantages of the proposed method are verified by numerical simulations. The average signal to noise ratio of the EMT system is 48 dB. Experimental results verified that the reconstructed images can show the number and positions of the imaging objects, demonstrating the feasibility and effectiveness of the novel imaging system design.

BiFSMNv2: Pushing Binary Neural Networks for Keyword Spotting to Real-Network Performance.

Deep neural networks, such as the deep-FSMN, have been widely studied for keyword spotting (KWS) applications while suffering expensive computation and storage. Therefore, network compression technologies such as binarization are studied to deploy KWS models on edge. In this article, we present a strong yet efficient binary neural network for KWS, namely, BiFSMNv2, pushing it to the real-network accuracy performance. First, we present a dual-scale thinnable 1-bit-architecture (DTA) to recover the representation capability of the binarized computation units by dual-scale activation binarization and liberate the speedup potential from an overall architecture perspective. Second, we also construct a frequency-independent distillation (FID) scheme for KWS binarization-aware training, which distills the high-and low-frequency components independently to mitigate the information mismatch between full-precision and binarized representations. Moreover, we propose the learning propagation binarizer (LPB), a general and efficient binarizer that enables the forward and backward propagation of binary KWS networks to be continuously improved through learning. We implement and deploy BiFSMNv2 on ARMv8 real-world hardware with a novel fast bitwise computation kernel (FBCK), which is proposed to fully use registers and increase instruction throughput. Comprehensive experiments show our BiFSMNv2 outperforms the existing binary networks for KWS by convincing margins across different datasets and achieves comparable accuracy with the full-precision networks (only a tiny 1.51% drop on Speech Commands V1-12). We highlight that benefiting from the compact architecture and optimized hardware kernel, BiFSMNv2 can achieve an impressive 25.1 × speedup and 20.2 × storage-saving on edge hardware.

Spatiotemporal trends in neonatal, infant, and child mortality (1990-2019) based on Bayesian spatiotemporal modeling.

Background: Neonatal mortality rate (NMR), infant mortality rate (IMR), and child mortality rate (CMR) show a huge difference across countries, which has been posing challenges for public health policies and medical resource allocation. Methods: Bayesian spatiotemporal model is applied to assess the detailed spatiotemporal evolution of NMR, IMR, and CMR from a global perspective. Panel data from 185 countries from 1990 to 2019 are collected. Results: The continuously decreasing trend of NMR, IMR, and CMR indicated a great improvement in neonatal, infant, and child mortality worldwide. Further, huge differences in the NMR, IMR, and CMR still exist across countries. In addition, the gap of NMR, IMR, and CMR across the countries presented a widening trend from the perspective of dispersion degree and kernel densities. The spatiotemporal heterogeneities demonstrated that the decline degree among these three indicators could be observed as CMR > IMR > NMR. Countries such as Brazil, Sweden, Libya, Myanmar, Thailand, Uzbekistan, Greece, and Zimbabwe showed the highest values of b1i , indicating a weaker downward trend compared to the overall downward trend in the world. Conclusions: This study revealed the spatiotemporal patterns and trends in the levels and improvement of NMR, IMR, and CMR across countries. Further, NMR, IMR, and CMR show a continuously decreasing trend, but the differences in improvement degree present a widening trend across countries. This study provides further implications for policy in newborns, infants, and children's health to reduce health inequality worldwide.

Regionwise Generative Adversarial Image Inpainting for Large Missing Areas.

Recently, deep neural networks have achieved promising performance for in-filling large missing regions in image inpainting tasks. They have usually adopted the standard convolutional architecture over the corrupted image, leading to meaningless contents, such as color discrepancy, blur, and other artifacts. Moreover, most inpainting approaches cannot handle well the case of a large contiguous missing area. To address these problems, we propose a generic inpainting framework capable of handling incomplete images with both contiguous and discontiguous large missing areas. We pose this in an adversarial manner, deploying regionwise operations in both the generator and discriminator to separately handle the different types of regions, namely, existing regions and missing ones. Moreover, a correlation loss is introduced to capture the nonlocal correlations between different patches, and thus, guide the generator to obtain more information during inference. With the help of regionwise generative adversarial mechanism, our framework can restore semantically reasonable and visually realistic images for both discontiguous and contiguous large missing areas. Extensive experiments on three widely used datasets for image inpainting task have been conducted, and both qualitative and quantitative experimental results demonstrate that the proposed model significantly outperforms the state-of-the-art approaches, on the large contiguous and discontiguous missing areas.

Identification of copper metabolism and cuproptosis-related subtypes for predicting prognosis tumor microenvironment and drug candidates in hepatocellular carcinoma.

Copper (Cu) is an essential element of organisms, which can affect the survival of cells. However, the role of copper metabolism and cuproptosis on hepatic carcinoma is still unclear. In this study, the TCGA database was used as the test set, and the ICGC database and self-built database were used as the validation set. We screened out a class of copper metabolism and cuproptosis-related genes (CMCRGs) that could influence hepatic carcinoma prognosis by survival analysis and differential comparison. Based on CMCRGs, patients were divided into two subtypes by cluster analysis. The C2 subtype was defined as the high copper related subtype, while the C1 subtype was defied as the low copper related subtype. At the clinical level, compared with the C1 subtype, the C2 subtype had higher grade pathological features, risk scores, and worse survival. In addition, the immune response and metabolic status also differed between C1 and C2. Specifically, C2 subtype had a higher proportion of immune cell composition and highly expressed immune checkpoint genes. C2 subtype had a higher TIDE score with a higher proportion of tumor immune dysfunction and exclusion. At the molecular level, the C2 subtype had a higher frequency of driver gene mutations (TP53 and OBSCN). Mechanistically, the single nucleotide polymorphisms of C2 subtype had a very strong transcriptional strand bias for C>A mutations. Copy number variations in the C2 subtype were characterized by LOXL3 CNV gain, which also showed high association with PDCD1/CTLA4. Finally, drug sensitivity responsiveness was assessed in both subtypes. C2 subtype had lower IC50 values for targeted and chemotherapeutic agents (sorafenib, imatinib and methotrexate, etc.). Thus, CMCRGs related subtypes showed poor response to immunotherapy and better responsiveness to targeted agents, and the results might provide a reference for precision treatment of hepatic carcinoma.

Integrated 3d flow-based multi-atlas brain structure segmentation.

MRI brain structure segmentation plays an important role in neuroimaging studies. Existing methods either spend much CPU time, require considerable annotated data, or fail in segmenting volumes with large deformation. In this paper, we develop a novel multi-atlas-based algorithm for 3D MRI brain structure segmentation. It consists of three modules: registration, atlas selection and label fusion. Both registration and label fusion leverage an integrated flow based on grayscale and SIFT features. We introduce an effective and efficient strategy for atlas selection by employing the accompanying energy generated in the registration step. A 3D sequential belief propagation method and a 3D coarse-to-fine flow matching approach are developed in both registration and label fusion modules. The proposed method is evaluated on five public datasets. The results show that it has the best performance in almost all the settings compared to competitive methods such as ANTs, Elastix, Learning to Rank and Joint Label Fusion. Moreover, our registration method is more than 7 times as efficient as that of ANTs SyN, while our label transfer method is 18 times faster than Joint Label Fusion in CPU time. The results on the ADNI dataset demonstrate that our method is applicable to image pairs that require a significant transformation in registration. The performance on a composite dataset suggests that our method succeeds in a cross-modality manner. The results of this study show that the integrated 3D flow-based method is effective and efficient for brain structure segmentation. It also demonstrates the power of SIFT features, multi-atlas segmentation and classical machine learning algorithms for a medical image analysis task. The experimental results on public datasets show the proposed method's potential for general applicability in various brain structures and settings.

Classifying early stages of cervical cancer with MRI-based radiomics.

This study aims to establish a MRI-based classifier to distinguish early stages of cervical cancer with improved diagnostic performance to assist clinical diagnosis and treatment. 57 patients with pathological diagnosis of cervical cancer from January 2018 to May 2019 were enrolled in this study. MRI examinations, including T1-weighted image(T1WI), T2-weighted image(T2W), diffusion weighted imaging (DWI) and dynamic contrast enhanced (DCE), were performed before surgery. MR images from patients of stage Ib or IIa cervical cancer with tumor segmented were used as input. Feature extraction process extracted first-order statistics and texture and applied filters. The dimensionality of the radiomic features was reduced using the least absolute shrinkage and selection operator (LASSO). Models were trained by three machine-learning (k-nearest neighbor (KNN), support vector machine (SVM), and logistic regression (LR)) and diagnostic performance in differentiating stage Ib and stage IIa cases was evaluated. A total of 27 features were extracted to establish models, including 2 features from T1WI, 5 features from T2WI, 5 features from DWI (b = 50), 4 features from DWI (b = 800), 5 features from DCE, and 6 features from ADC. For each machine learning (ML) classifier, six sequences of training set and testing set are modeled and analyzed. Among all the models, the training set and testing set of T2WI model built by SVM classifier were the best (Area under the curve (AUC) 0.915) / (AUC 0.907). Radiomic analysis of ML-based texture features and first-order statistics features can be used to stage the early cervical cancer pre-operatively.

Transductive Relation-Propagation With Decoupling Training for Few-Shot Learning.

Few-shot learning, aiming to learn novel concepts from one or a few labeled examples, is an interesting and very challenging problem with many practical advantages. Existing few-shot methods usually utilize data of the same classes to train the feature embedding module and in a row, which is unable to learn adapting to new tasks. Besides, traditional few-shot models fail to take advantage of the valuable relations of the support-query pairs, leading to performance degradation. In this article, we propose a transductive relation-propagation graph neural network (GNN) with a decoupling training strategy (TRPN-D) to explicitly model and propagate such relations across support-query pairs, and empower the few-shot module the ability of transferring past knowledge to new tasks via the decoupling training. Our few-shot module, namely TRPN, treats the relation of each support-query pair as a graph node, named relational node, and resorts to the known relations between support samples, including both intraclass commonality and interclass uniqueness. Through relation propagation, the model could generate the discriminative relation embeddings for support-query pairs. To the best of our knowledge, this is the first work that decouples the training of the embedding network and the few-shot graph module with different tasks, which might offer a new way to solve the few-shot learning problem. Extensive experiments conducted on several benchmark datasets demonstrate that our method can significantly outperform a variety of state-of-the-art few-shot learning methods.

Universal Adversarial Patch Attack for Automatic Checkout Using Perceptual and Attentional Bias.

Adversarial examples are inputs with imperceptible perturbations that easily mislead deep neural networks (DNNs). Recently, adversarial patch, with noise confined to a small and localized patch, has emerged for its easy feasibility in real-world scenarios. However, existing strategies failed to generate adversarial patches with strong generalization ability due to the ignorance of the inherent biases of models. In other words, the adversarial patches are always input-specific and fail to attack images from all classes or different models, especially unseen classes and black-box models. To address the problem, this paper proposes a bias-based framework to generate universal adversarial patches with strong generalization ability, which exploits the perceptual bias and attentional bias to improve the attacking ability. Regarding the perceptual bias, since DNNs are strongly biased towards textures, we exploit the hard examples which convey strong model uncertainties and extract a textural patch prior from them by adopting the style similarities. The patch prior is closer to decision boundaries and would promote attacks across classes. As for the attentional bias, motivated by the fact that different models share similar attention patterns towards the same image, we exploit this bias by confusing the model-shared similar attention patterns. Thus, the generated adversarial patches can obtain stronger transferability among different models. Taking Automatic Check-out (ACO) as the typical scenario, extensive experiments including white-box/black-box settings in both digital-world (RPC, the largest ACO related dataset) and physical-world scenario (Taobao and JD, the world's largest online shopping platforms) are conducted. Experimental results demonstrate that our proposed framework outperforms state-of-the-art adversarial patch attack methods.